Observability Economics: DataDog Price Shocks, OpenTelemetry Migration, and the Tail Cost of Instrumentation

Observability, the category that covers logs, metrics, and traces, has spent the last three years quietly moving from an engineering line item to a material item on the technology P&L. The pattern is the same across most of the mid-cap and large US technology organizations we work with: instrumentation was added generously through the 2020-2023 build-out, the volume of emitted telemetry grew faster than any budget model anticipated, and the annual bills from the major observability vendors began to attract the attention of the CFO in a way they had not before. The response through 2025 and into 2026 has been serious, and it is worth describing precisely because most of the public commentary on the DataDog price shocks missed the more consequential structural shift underneath.

The first thing that changed is that CTOs stopped treating observability as a single vendor decision. The pattern we now see routinely is a portfolio: a primary vendor for the workloads where the vendor's product genuinely adds value, an open-source or self-managed stack for the high-volume, low-differentiation telemetry that was previously going into the same bill, and, increasingly, an OpenTelemetry-based collection layer that keeps the option to change destinations open. The specific split varies by organization, but the principle is that not every log, metric, and trace deserves the premium price of a fully-managed observability platform, and separating the two has produced substantial savings without meaningfully degrading the operational posture.

The OpenTelemetry migration is the underlying enabler. OpenTelemetry, as a standard for collecting and exporting telemetry data, has matured to the point where the collection layer is genuinely vendor-neutral. That means the historical lock-in of the observability vendors, which came from proprietary agents deeply embedded in application code, is being systematically dismantled. Organizations that have moved their collection layer to OpenTelemetry have discovered that the switching cost between destinations, whether commercial or open-source, is now a real option rather than a theoretical one. The vendors know this, and the pricing conversation with them has changed accordingly.

The second structural change is on sampling and retention. The default posture during the build-out era was to collect everything and retain it for as long as the vendor's tier allowed. The current posture, in mature programs, is that not all telemetry is equally valuable, and the discipline of deciding what to sample at head, what to sample at tail, and what to drop entirely has moved from a specialist concern to a mainstream operational choice. Head-sampling of high-volume, low-cardinality traces, tail-sampling for the transactions that actually failed or exceeded latency thresholds, and short-retention for logs that no one has actually queried in the last quarter, are the specific levers being pulled. The savings are real, and the operational blindness that was feared largely has not materialized, because the material incidents were rarely diagnosed from the deleted data anyway.

The third change is more organizational. Observability is no longer an engineering-only budget. In the mature programs, the platform team owns the collection layer and the destination contracts, but individual application teams see and are accountable for their own emitted volume. The chargeback or showback model, whichever the organization prefers, makes the cost of adding a debug log at production log level visible to the team adding it. That accountability has done more to reduce the growth curve than any single vendor negotiation, because the incentive to instrument responsibly is now aligned with the incentive that actually pays the bill.

The FinOps interaction is worth naming. Observability spending has become one of the specific line items the FinOps function is expected to have a view on, alongside compute, storage, and increasingly AI inference. The organizations that have integrated observability into their FinOps discipline are the ones that are catching pricing tier changes at renewal rather than at the next quarter's variance review.

For a CTO in 2026, the working guidance is that if the observability bill is a top-ten line in the technology budget and there is not a documented plan for the collection layer, the sampling posture, and the vendor-portfolio strategy, the next renewal cycle will be materially harder than the last one. The specific levers, OpenTelemetry-based collection, per-team accountability, deliberate sampling, and open-source alternatives for the high-volume tail, are well-understood at this point and are producing savings in a range that comfortably justifies the engineering investment required to pull them. The organizations still on a single-vendor, everything-collected, indefinite-retention posture are the ones the CFO is starting to ask questions about, and the answer the CTO needs to be ready with is the plan, not the bill.